fs/nfsd/nfssvc.c

   1 /*
   2  * linux/fs/nfsd/nfssvc.c
   3  *
   4  * Central processing for nfsd.
   5  *
   6  * Authors:     Olaf Kirch (okir@monad.swb.de)
   7  *
   8  * Copyright (C) 1995, 1996, 1997 Olaf Kirch <okir@monad.swb.de>
   9  */
  10
  11 #include <linux/module.h>
  12 #include <linux/sched.h>
  13 #include <linux/time.h>
  14 #include <linux/errno.h>
  15 #include <linux/nfs.h>
  16 #include <linux/in.h>
  17 #include <linux/uio.h>
  18 #include <linux/unistd.h>
  19 #include <linux/slab.h>
  20 #include <linux/smp.h>
  21 #include <linux/smp_lock.h>
  22 #include <linux/freezer.h>
  23 #include <linux/fs_struct.h>
  24 #include <linux/kthread.h>
  25 #include <linux/swap.h>
  26
  27 #include <linux/sunrpc/types.h>
  28 #include <linux/sunrpc/stats.h>
  29 #include <linux/sunrpc/svc.h>
  30 #include <linux/sunrpc/svcsock.h>
  31 #include <linux/sunrpc/cache.h>
  32 #include <linux/nfsd/nfsd.h>
  33 #include <linux/nfsd/stats.h>
  34 #include <linux/nfsd/cache.h>
  35 #include <linux/nfsd/syscall.h>
  36 #include <linux/lockd/bind.h>
  37 #include <linux/nfsacl.h>
  38
  39 #define NFSDDBG_FACILITY        NFSDDBG_SVC
  40
  41 extern struct svc_program       nfsd_program;
  42 static int                      nfsd(void *vrqstp);
  43 struct timeval                  nfssvc_boot;
  44
  45 /*
  46  * nfsd_mutex protects nfsd_serv -- both the pointer itself and the members
  47  * of the svc_serv struct. In particular, ->sv_nrthreads but also to some
  48  * extent ->sv_temp_socks and ->sv_permsocks. It also protects nfsdstats.th_cnt
  49  *
  50  * If (out side the lock) nfsd_serv is non-NULL, then it must point to a
  51  * properly initialised 'struct svc_serv' with ->sv_nrthreads > 0. That number
  52  * of nfsd threads must exist and each must listed in ->sp_all_threads in each
  53  * entry of ->sv_pools[].
  54  *
  55  * Transitions of the thread count between zero and non-zero are of particular
  56  * interest since the svc_serv needs to be created and initialized at that
  57  * point, or freed.
  58  *
  59  * Finally, the nfsd_mutex also protects some of the global variables that are
  60  * accessed when nfsd starts and that are settable via the write_* routines in
  61  * nfsctl.c. In particular:
  62  *
  63  *      user_recovery_dirname
  64  *      user_lease_time
  65  *      nfsd_versions
  66  */
  67 DEFINE_MUTEX(nfsd_mutex);
  68 struct svc_serv                 *nfsd_serv;
  69
  70 /*
  71  * nfsd_drc_lock protects nfsd_drc_max_pages and nfsd_drc_pages_used.
  72  * nfsd_drc_max_pages limits the total amount of memory available for
  73  * version 4.1 DRC caches.
  74  * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
  75  */
  76 spinlock_t      nfsd_drc_lock;
  77 unsigned int    nfsd_drc_max_pages;
  78 unsigned int    nfsd_drc_pages_used;
  79
  80 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
  81 static struct svc_stat  nfsd_acl_svcstats;
  82 static struct svc_version *     nfsd_acl_version[] = {
  83         [2] = &nfsd_acl_version2,
  84         [3] = &nfsd_acl_version3,
  85 };
  86
  87 #define NFSD_ACL_MINVERS            2
  88 #define NFSD_ACL_NRVERS         ARRAY_SIZE(nfsd_acl_version)
  89 static struct svc_version *nfsd_acl_versions[NFSD_ACL_NRVERS];
  90
  91 static struct svc_program       nfsd_acl_program = {
  92         .pg_prog                = NFS_ACL_PROGRAM,
  93         .pg_nvers               = NFSD_ACL_NRVERS,
  94         .pg_vers                = nfsd_acl_versions,
  95         .pg_name                = "nfsacl",
  96         .pg_class               = "nfsd",
  97         .pg_stats               = &nfsd_acl_svcstats,
  98         .pg_authenticate        = &svc_set_client,
  99 };
 100
 101 static struct svc_stat  nfsd_acl_svcstats = {
 102         .program        = &nfsd_acl_program,
 103 };
 104 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */
 105
 106 static struct svc_version *     nfsd_version[] = {
 107         [2] = &nfsd_version2,
 108 #if defined(CONFIG_NFSD_V3)
 109         [3] = &nfsd_version3,
 110 #endif
 111 #if defined(CONFIG_NFSD_V4)
 112         [4] = &nfsd_version4,
 113 #endif
 114 };
 115
 116 #define NFSD_MINVERS            2
 117 #define NFSD_NRVERS             ARRAY_SIZE(nfsd_version)
 118 static struct svc_version *nfsd_versions[NFSD_NRVERS];
 119
 120 struct svc_program              nfsd_program = {
 121 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 122         .pg_next                = &nfsd_acl_program,
 123 #endif
 124         .pg_prog                = NFS_PROGRAM,          /* program number */
 125         .pg_nvers               = NFSD_NRVERS,          /* nr of entries in nfsd_version */
 126         .pg_vers                = nfsd_versions,        /* version table */
 127         .pg_name                = "nfsd",               /* program name */
 128         .pg_class               = "nfsd",               /* authentication class */
 129         .pg_stats               = &nfsd_svcstats,       /* version table */
 130         .pg_authenticate        = &svc_set_client,      /* export authentication */
 131
 132 };
 133
 134 u32 nfsd_supported_minorversion;
 135
 136 int nfsd_vers(int vers, enum vers_op change)
 137 {
 138         if (vers < NFSD_MINVERS || vers >= NFSD_NRVERS)
 139                 return -1;
 140         switch(change) {
 141         case NFSD_SET:
 142                 nfsd_versions[vers] = nfsd_version[vers];
 143 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 144                 if (vers < NFSD_ACL_NRVERS)
 145                         nfsd_acl_versions[vers] = nfsd_acl_version[vers];
 146 #endif
 147                 break;
 148         case NFSD_CLEAR:
 149                 nfsd_versions[vers] = NULL;
 150 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 151                 if (vers < NFSD_ACL_NRVERS)
 152                         nfsd_acl_versions[vers] = NULL;
 153 #endif
 154                 break;
 155         case NFSD_TEST:
 156                 return nfsd_versions[vers] != NULL;
 157         case NFSD_AVAIL:
 158                 return nfsd_version[vers] != NULL;
 159         }
 160         return 0;
 161 }
 162
 163 int nfsd_minorversion(u32 minorversion, enum vers_op change)
 164 {
 165         if (minorversion > NFSD_SUPPORTED_MINOR_VERSION)
 166                 return -1;
 167         switch(change) {
 168         case NFSD_SET:
 169                 nfsd_supported_minorversion = minorversion;
 170                 break;
 171         case NFSD_CLEAR:
 172                 if (minorversion == 0)
 173                         return -1;
 174                 nfsd_supported_minorversion = minorversion - 1;
 175                 break;
 176         case NFSD_TEST:
 177                 return minorversion <= nfsd_supported_minorversion;
 178         case NFSD_AVAIL:
 179                 return minorversion <= NFSD_SUPPORTED_MINOR_VERSION;
 180         }
 181         return 0;
 182 }
 183
 184 /*
 185  * Maximum number of nfsd processes
 186  */
 187 #define NFSD_MAXSERVS           8192
 188
 189 int nfsd_nrthreads(void)
 190 {
 191         int rv = 0;
 192         mutex_lock(&nfsd_mutex);
 193         if (nfsd_serv)
 194                 rv = nfsd_serv->sv_nrthreads;
 195         mutex_unlock(&nfsd_mutex);
 196         return rv;
 197 }
 198
 199 static void nfsd_last_thread(struct svc_serv *serv)
 200 {
 201         /* When last nfsd thread exits we need to do some clean-up */
 202         struct svc_xprt *xprt;
 203         list_for_each_entry(xprt, &serv->sv_permsocks, xpt_list)
 204                 lockd_down();
 205         nfsd_serv = NULL;
 206         nfsd_racache_shutdown();
 207         nfs4_state_shutdown();
 208
 209         printk(KERN_WARNING "nfsd: last server has exited, flushing export "
 210                             "cache\n");
 211         nfsd_export_flush();
 212 }
 213
 214 void nfsd_reset_versions(void)
 215 {
 216         int found_one = 0;
 217         int i;
 218
 219         for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++) {
 220                 if (nfsd_program.pg_vers[i])
 221                         found_one = 1;
 222         }
 223
 224         if (!found_one) {
 225                 for (i = NFSD_MINVERS; i < NFSD_NRVERS; i++)
 226                         nfsd_program.pg_vers[i] = nfsd_version[i];
 227 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
 228                 for (i = NFSD_ACL_MINVERS; i < NFSD_ACL_NRVERS; i++)
 229                         nfsd_acl_program.pg_vers[i] =
 230                                 nfsd_acl_version[i];
 231 #endif
 232         }
 233 }
 234
 235 /*
 236  * Each session guarantees a negotiated per slot memory cache for replies
 237  * which in turn consumes memory beyond the v2/v3/v4.0 server. A dedicated
 238  * NFSv4.1 server might want to use more memory for a DRC than a machine
 239  * with mutiple services.
 240  *
 241  * Impose a hard limit on the number of pages for the DRC which varies
 242  * according to the machines free pages. This is of course only a default.
 243  *
 244  * For now this is a #defined shift which could be under admin control
 245  * in the future.
 246  */
 247 static void set_max_drc(void)
 248 {
 249         /* The percent of nr_free_buffer_pages used by the V4.1 server DRC */
 250         #define NFSD_DRC_SIZE_SHIFT     7
 251         nfsd_drc_max_pages = nr_free_buffer_pages()
 252                                                 >> NFSD_DRC_SIZE_SHIFT;
 253         nfsd_drc_pages_used = 0;
 254         spin_lock_init(&nfsd_drc_lock);
 255         dprintk("%s nfsd_drc_max_pages %u\n", __func__,
 256                 nfsd_drc_max_pages);
 257 }
 258
 259 int nfsd_create_serv(void)
 260 {
 261         int err = 0;
 262
 263         WARN_ON(!mutex_is_locked(&nfsd_mutex));
 264         if (nfsd_serv) {
 265                 svc_get(nfsd_serv);
 266                 return 0;
 267         }
 268         if (nfsd_max_blksize == 0) {
 269                 /* choose a suitable default */
 270                 struct sysinfo i;
 271                 si_meminfo(&i);
 272                 /* Aim for 1/4096 of memory per thread
 273                  * This gives 1MB on 4Gig machines
 274                  * But only uses 32K on 128M machines.
 275                  * Bottom out at 8K on 32M and smaller.
 276                  * Of course, this is only a default.
 277                  */
 278                 nfsd_max_blksize = NFSSVC_MAXBLKSIZE;
 279                 i.totalram <<= PAGE_SHIFT - 12;
 280                 while (nfsd_max_blksize > i.totalram &&
 281                        nfsd_max_blksize >= 8*1024*2)
 282                         nfsd_max_blksize /= 2;
 283         }
 284
 285         nfsd_serv = svc_create_pooled(&nfsd_program, nfsd_max_blksize,
 286                                       nfsd_last_thread, nfsd, THIS_MODULE);
 287         if (nfsd_serv == NULL)
 288                 err = -ENOMEM;
 289         else
 290                 set_max_drc();
 291
 292         do_gettimeofday(&nfssvc_boot);          /* record boot time */
 293         return err;
 294 }
 295
 296 static int nfsd_init_socks(int port)
 297 {
 298         int error;
 299         if (!list_empty(&nfsd_serv->sv_permsocks))
 300                 return 0;
 301
 302         error = svc_create_xprt(nfsd_serv, "udp", PF_INET, port,
 303                                         SVC_SOCK_DEFAULTS);
 304         if (error < 0)
 305                 return error;
 306
 307         error = lockd_up();
 308         if (error < 0)
 309                 return error;
 310
 311         error = svc_create_xprt(nfsd_serv, "tcp", PF_INET, port,
 312                                         SVC_SOCK_DEFAULTS);
 313         if (error < 0)
 314                 return error;
 315
 316         error = lockd_up();
 317         if (error < 0)
 318                 return error;
 319
 320         return 0;
 321 }
 322
 323 int nfsd_nrpools(void)
 324 {
 325         if (nfsd_serv == NULL)
 326                 return 0;
 327         else
 328                 return nfsd_serv->sv_nrpools;
 329 }
 330
 331 int nfsd_get_nrthreads(int n, int *nthreads)
 332 {
 333         int i = 0;
 334
 335         if (nfsd_serv != NULL) {
 336                 for (i = 0; i < nfsd_serv->sv_nrpools && i < n; i++)
 337                         nthreads[i] = nfsd_serv->sv_pools[i].sp_nrthreads;
 338         }
 339
 340         return 0;
 341 }
 342
 343 int nfsd_set_nrthreads(int n, int *nthreads)
 344 {
 345         int i = 0;
 346         int tot = 0;
 347         int err = 0;
 348
 349         WARN_ON(!mutex_is_locked(&nfsd_mutex));
 350
 351         if (nfsd_serv == NULL || n <= 0)
 352                 return 0;
 353
 354         if (n > nfsd_serv->sv_nrpools)
 355                 n = nfsd_serv->sv_nrpools;
 356
 357         /* enforce a global maximum number of threads */
 358         tot = 0;
 359         for (i = 0; i < n; i++) {
 360                 if (nthreads[i] > NFSD_MAXSERVS)
 361                         nthreads[i] = NFSD_MAXSERVS;
 362                 tot += nthreads[i];
 363         }
 364         if (tot > NFSD_MAXSERVS) {
 365                 /* total too large: scale down requested numbers */
 366                 for (i = 0; i < n && tot > 0; i++) {
 367                         int new = nthreads[i] * NFSD_MAXSERVS / tot;
 368                         tot -= (nthreads[i] - new);
 369                         nthreads[i] = new;
 370                 }
 371                 for (i = 0; i < n && tot > 0; i++) {
 372                         nthreads[i]--;
 373                         tot--;
 374                 }
 375         }
 376
 377         /*
 378          * There must always be a thread in pool 0; the admin
 379          * can't shut down NFS completely using pool_threads.
 380          */
 381         if (nthreads[0] == 0)
 382                 nthreads[0] = 1;
 383
 384         /* apply the new numbers */
 385         svc_get(nfsd_serv);
 386         for (i = 0; i < n; i++) {
 387                 err = svc_set_num_threads(nfsd_serv, &nfsd_serv->sv_pools[i],
 388                                           nthreads[i]);
 389                 if (err)
 390                         break;
 391         }
 392         svc_destroy(nfsd_serv);
 393
 394         return err;
 395 }
 396
 397 int
 398 nfsd_svc(unsigned short port, int nrservs)
 399 {
 400         int     error;
 401
 402         mutex_lock(&nfsd_mutex);
 403         dprintk("nfsd: creating service\n");
 404         if (nrservs <= 0)
 405                 nrservs = 0;
 406         if (nrservs > NFSD_MAXSERVS)
 407                 nrservs = NFSD_MAXSERVS;
 408         error = 0;
 409         if (nrservs == 0 && nfsd_serv == NULL)
 410                 goto out;
 411
 412         /* Readahead param cache - will no-op if it already exists */
 413         error = nfsd_racache_init(2*nrservs);
 414         if (error<0)
 415                 goto out;
 416         nfs4_state_start();
 417
 418         nfsd_reset_versions();
 419
 420         error = nfsd_create_serv();
 421
 422         if (error)
 423                 goto out;
 424         error = nfsd_init_socks(port);
 425         if (error)
 426                 goto failure;
 427
 428         error = svc_set_num_threads(nfsd_serv, NULL, nrservs);
 429         if (error == 0)
 430                 /* We are holding a reference to nfsd_serv which
 431                  * we don't want to count in the return value,
 432                  * so subtract 1
 433                  */
 434                 error = nfsd_serv->sv_nrthreads - 1;
 435  failure:
 436         svc_destroy(nfsd_serv);         /* Release server */
 437  out:
 438         mutex_unlock(&nfsd_mutex);
 439         return error;
 440 }
 441
 442
 443 /*
 444  * This is the NFS server kernel thread
 445  */
 446 static int
 447 nfsd(void *vrqstp)
 448 {
 449         struct svc_rqst *rqstp = (struct svc_rqst *) vrqstp;
 450         int err, preverr = 0;
 451
 452         /* Lock module and set up kernel thread */
 453         mutex_lock(&nfsd_mutex);
 454
 455         /* At this point, the thread shares current->fs
 456          * with the init process. We need to create files with a
 457          * umask of 0 instead of init's umask. */
 458         if (unshare_fs_struct() < 0) {
 459                 printk("Unable to start nfsd thread: out of memory\n");
 460                 goto out;
 461         }
 462
 463         current->fs->umask = 0;
 464
 465         /*
 466          * thread is spawned with all signals set to SIG_IGN, re-enable
 467          * the ones that will bring down the thread
 468          */
 469         allow_signal(SIGKILL);
 470         allow_signal(SIGHUP);
 471         allow_signal(SIGINT);
 472         allow_signal(SIGQUIT);
 473
 474         nfsdstats.th_cnt++;
 475         mutex_unlock(&nfsd_mutex);
 476
 477         /*
 478          * We want less throttling in balance_dirty_pages() so that nfs to
 479          * localhost doesn't cause nfsd to lock up due to all the client's
 480          * dirty pages.
 481          */
 482         current->flags |= PF_LESS_THROTTLE;
 483         set_freezable();
 484
 485         /*
 486          * The main request loop
 487          */
 488         for (;;) {
 489                 /*
 490                  * Find a socket with data available and call its
 491                  * recvfrom routine.
 492                  */
 493                 while ((err = svc_recv(rqstp, 60*60*HZ)) == -EAGAIN)
 494                         ;
 495                 if (err == -EINTR)
 496                         break;
 497                 else if (err < 0) {
 498                         if (err != preverr) {
 499                                 printk(KERN_WARNING "%s: unexpected error "
 500                                         "from svc_recv (%d)\n", __func__, -err);
 501                                 preverr = err;
 502                         }
 503                         schedule_timeout_uninterruptible(HZ);
 504                         continue;
 505                 }
 506
 507
 508                 /* Lock the export hash tables for reading. */
 509                 exp_readlock();
 510
 511                 svc_process(rqstp);
 512
 513                 /* Unlock export hash tables */
 514                 exp_readunlock();
 515         }
 516
 517         /* Clear signals before calling svc_exit_thread() */
 518         flush_signals(current);
 519
 520         mutex_lock(&nfsd_mutex);
 521         nfsdstats.th_cnt --;
 522
 523 out:
 524         /* Release the thread */
 525         svc_exit_thread(rqstp);
 526
 527         /* Release module */
 528         mutex_unlock(&nfsd_mutex);
 529         module_put_and_exit(0);
 530         return 0;
 531 }
 532
 533 static __be32 map_new_errors(u32 vers, __be32 nfserr)
 534 {
 535         if (nfserr == nfserr_jukebox && vers == 2)
 536                 return nfserr_dropit;
 537         if (nfserr == nfserr_wrongsec && vers < 4)
 538                 return nfserr_acces;
 539         return nfserr;
 540 }
 541
 542 int
 543 nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
 544 {
 545         struct svc_procedure    *proc;
 546         kxdrproc_t              xdr;
 547         __be32                  nfserr;
 548         __be32                  *nfserrp;
 549
 550         dprintk("nfsd_dispatch: vers %d proc %d\n",
 551                                 rqstp->rq_vers, rqstp->rq_proc);
 552         proc = rqstp->rq_procinfo;
 553
 554         /* Check whether we have this call in the cache. */
 555         switch (nfsd_cache_lookup(rqstp, proc->pc_cachetype)) {
 556         case RC_INTR:
 557         case RC_DROPIT:
 558                 return 0;
 559         case RC_REPLY:
 560                 return 1;
 561         case RC_DOIT:;
 562                 /* do it */
 563         }
 564
 565         /* Decode arguments */
 566         xdr = proc->pc_decode;
 567         if (xdr && !xdr(rqstp, (__be32*)rqstp->rq_arg.head[0].iov_base,
 568                         rqstp->rq_argp)) {
 569                 dprintk("nfsd: failed to decode arguments!\n");
 570                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 571                 *statp = rpc_garbage_args;
 572                 return 1;
 573         }
 574
 575         /* need to grab the location to store the status, as
 576          * nfsv4 does some encoding while processing
 577          */
 578         nfserrp = rqstp->rq_res.head[0].iov_base
 579                 + rqstp->rq_res.head[0].iov_len;
 580         rqstp->rq_res.head[0].iov_len += sizeof(__be32);
 581
 582         /* NFSv4.1 DRC requires statp */
 583         if (rqstp->rq_vers == 4)
 584                 nfsd4_set_statp(rqstp, statp);
 585
 586         /* Now call the procedure handler, and encode NFS status. */
 587         nfserr = proc->pc_func(rqstp, rqstp->rq_argp, rqstp->rq_resp);
 588         nfserr = map_new_errors(rqstp->rq_vers, nfserr);
 589         if (nfserr == nfserr_dropit) {
 590                 dprintk("nfsd: Dropping request; may be revisited later\n");
 591                 nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 592                 return 0;
 593         }
 594
 595         if (rqstp->rq_proc != 0)
 596                 *nfserrp++ = nfserr;
 597
 598         /* Encode result.
 599          * For NFSv2, additional info is never returned in case of an error.
 600          */
 601         if (!(nfserr && rqstp->rq_vers == 2)) {
 602                 xdr = proc->pc_encode;
 603                 if (xdr && !xdr(rqstp, nfserrp,
 604                                 rqstp->rq_resp)) {
 605                         /* Failed to encode result. Release cache entry */
 606                         dprintk("nfsd: failed to encode result!\n");
 607                         nfsd_cache_update(rqstp, RC_NOCACHE, NULL);
 608                         *statp = rpc_system_err;
 609                         return 1;
 610                 }
 611         }
 612
 613         /* Store reply in cache. */
 614         nfsd_cache_update(rqstp, proc->pc_cachetype, statp + 1);
 615         return 1;
 616 }
 617
 618 int nfsd_pool_stats_open(struct inode *inode, struct file *file)
 619 {
 620         if (nfsd_serv == NULL)
 621                 return -ENODEV;
 622         return svc_pool_stats_open(nfsd_serv, file);
 623 }